AI in Healthcare: Predictive Medicine


However trained we are in a concrete task, human beings make mistakes that machines would never allow. In addition, our senses have a limit, which prevents us from reaching where the robots can reach.

Therefore, the development of artificial intelligence is essential in a large number of areas, among which medicine could not be lacking. Thanks to it, diagnoses can be perfected in hospitals and even your home with the use of a smartphone or any wearable device to keep some parameters of your health under control.

All the advances achieved in this regard grow every day by leaps and bounds, obtaining milestones that just a few years ago would have seemed the result of science fiction. These are some of its most important applications:

Breast Cancer Screening

Artificial Intelligence tools do not replace medical professionals but provide an accurate second opinion.

Although doctors are usually well trained in the detection of tumors on mammograms, sometimes one of these images may have ambiguities, which can be detected more easily with the help of RPA. This is what led the 18-year-old adolescent Abu Qader to design an app that could detect possible abnormalities in mammograms more effectively than human beings.

In January 2018, a team of scientists from the Polytechnic University of Valencia, Spain (UPV), the Higher Council for Scientific Research (CSIC) and the University of Valencia (UV), among other institutions, developed a tool based on artificial intelligence, in order to reduce the number of false positives that usually occur with conventional methods. These mistakes are due to the fact that the techniques available so far focus on the most suspicious areas, and may lead to false alarms. On the contrary, artificial intelligence manages to provide information based on the possible presence of cancer, reducing the number of regions under suspicion. It is one of the most effective projects of this type so far since it manages to carry out its diagnoses with 90% success.

Other tools do not focus on the diagnosis of the tumor as such, but some risk factors, including dense breast tissue. This density is classified into four levels, from A, in which fatty tissue is abundant and dense is scarce, to D, where the opposite occurs completely. The last two are considered a risk factor for breast cancer. It is important to classify patients according to their breast density, to pay special attention to those who are at greater risk. The problem is that sometimes it becomes a subjective classification, which can lead to error. Therefore, last year scientists at MIT and the Massachusetts General Hospital designed a machine learning system capable of doing the same with higher exactitude. This is not intended to eliminate the role of the medical professional, but to provide a second opinion that would invite a second review if necessary.

Diagnosis of skin cancer

Skin cancer is the most frequent of all types of cancer. Therefore, the general population is recommended to prevent its appearance by taking appropriate precautions when exposed to the sun and also controlling the possible appearance of suspicious spots or moles. However, sometimes the waiting time to go to the dermatologist can generate great anguish, which would be solved if we could have an answer on the spot. For this reason, a team of scientists from Stanford University developed an application capable of diagnosing skin cancer through a photo taken with the mobile of the area in question.

To perform it, they trained a machine-learning algorithm with the help of 130,000 images of tumors of this type. Finally, the definitive result was tested by 21 dermatologists, obtaining very positive results.

Identification of rare genetic diseases

Not only tumors can be identified through an image, but also some rare genetic disorders, such as Angelman’s syndrome or Cornelia de Lange’s. Both diseases affect patients’ mobility or intellectual development. 

However, they also generate small changes in the physique that can be virtually imperceptible to humans, but not for a well-trained algorithm.

FDNA, a US-based company presented Deepgestalt, an application that can scan images of the faces of sick children and identify most of these syndromes, with 90% accuracy. In this case, they trained the algorithm with the help of 17,000 images, in which people with 200 different syndromes appeared. Not only was it possible to obtain diagnoses with great precision, but in the case of Noonan syndrome, it was able to identify which of the five possible genetic mutations that cause it occurred in each case. Of course, for this, the accuracy was reduced to 64%.

Those responsible for the study have warned that the application would not provide definitive evidence of the disease, as it should be finally confirmed with a genetic test, especially to identify the specific mutation that causes it. In any case, in the future, it could provide a first diagnosis much faster than offered by conventional methods.